Use an aggregation gateway instead of the Prometheus Pushgateway

[emschwartz]

The Pushgateway is meant for one-off batch jobs, not for long-running processes that keep updating the metrics. Metrics scraped from the Pushgateway will be incorrect when it's used with Autometrics.

We should use the prom-aggregation-gateway or the prometheus-gravel-gateway instead, both of which are built for aggregating metrics like we need for Autometrics.